Few questions about ddsp HOT 2 CLOSED

Sure,

1. sinusoidal_amps = amplitude * harmonic_distribution
   harmonic_distribution sums to 1 for every moment in time (thus I call it a distribution), so it distributes "amplitude" among the harmonics.

2. Yah, you can do inharmonicity, it's even an input to harmonic_synthesis(), but I just don't use it at the moment. I have some "pure sinusoidal" models, but haven't open sourced them yet.

3. 16kHz is just a standard in ML since audio ML is hard. We're working on extending it to 48kHz and almost have it working, and in principle it should be a lot easier than with other ML models, but there a couple hyperparameters to tweak. Also, personally, I kind of like the retro sound of lower bit rates (hides the model mistakes :). But yes, you can work at higher frequencies.

from ddsp.

Good stuff, many thanks!!

from ddsp.